HiNews
A new study published in GeoHealth has used machine learning to examine the factors that predict emergency room visits for mental and behavioral disorders (MBD) in youths across six cities in North Carolina, USA. The study found that socio-demographic variables, such as age, gender, and income, had a greater impact on youth mental health than environmental variables, such as temperature and vegetation. However, the study also revealed that different cities had different responses to environmental conditions, and that high temperatures might not always increase the risk of MBD in youths.
The researchers used data from over 42,000 ER visits for MBD in youths aged 5 to 24 during the summer months of 2016 to 2019. They paired the data with daily environmental and socio-demographic information to build four machine learning models: generalized linear model, generalized additive model, random forest, and extreme gradient boosting. They compared the performance of these models and used a method called SHapley Additive exPlanations (SHAP) to quantify the contribution of each variable to the prediction of MBD cases.
The results showed that the generalized additive model performed the best among the four models, and that socio-demographic variables were more important than environmental variables in predicting MBD cases at the aggregated city level. The most influential variables included the population of youths, the male-to-female ratio, and the median age of the city. The researchers also found that lower minimum temperatures, higher relative humidity, and lower vegetation greenness were associated with higher MBD cases.
However, when the researchers applied the generalized additive model to each individual city, they found no clear environmental variable that contributed to the highest risk of MBD. Instead, they found that different cities had different sensitivities to temperature changes. For example, in Charlotte and Raleigh, temperatures near the median were associated with higher MBD cases, while in Asheville and Wilmington, lower temperatures were associated with higher MBD cases.
The researchers suggested that these findings might reflect the adaptation of youths to different climatic regions and the availability of other resources that affect mental health. They also noted that their study was limited by the short study period and the lack of specific MBD diagnosis. They recommended further research to explore how air pollution, heat waves, and specific MBD types might interact with temperature and other factors to influence youth mental health.
The study is among the first to use machine learning to examine the driving factors behind MBD ER visits in youth in North Carolina. The study highlights the importance of local-level understanding and socio-demographic factors when trying to understand how temperature may influence MBD in youths. The study also provides new guidance on the application of machine learning models and SHAP values for predicting mental health conditions during high-temperature events.
